# State of AI Search Citations: Q4 2026

> A quarterly digest of published research on which sources AI search engines cite, how often answers change, and what the 2026 evidence overturned.

Source: https://metavert.io/state-of-ai-search-citations  
Published: 2026-10-07  
Updated: 2026-10-07

The **State of AI Search Citations** is a quarterly digest of published research on which sources AI answer engines cite, how stable those citations are, and which optimization claims survive testing. The short version for Q4 2026: citations have decoupled from organic rankings, engines rarely agree with one another, most citations go to third-party sites, and the on-page tactics promoted since 2023 have failed replication.

*Last updated: October 2026. This digest is refreshed quarterly.*

### Key Findings

- Only 37.9% of URLs cited in Google [AI Overviews](https://metavert.io/ai-overviews) rank in the organic top 10, down from about 76% in July 2025 (Ahrefs, March 2026); BrightEdge puts the overlap near 17% (February 2026).
- A re-test of the 2023 GEO levers on ten modern engine families found they "move citation on none" (Bajemon and Rochet, September 2026; vendor authors).
- Adding schema markup "produced no major uplift in citations on any platform" in a controlled study of 1,885 pages (Ahrefs, May 2026).
- Roughly 65% of the sources an engine cites for a query change from one day to the next (Schulte, Bleeker and Kaufmann, April 2026).
- Cross-platform source overlap is below 0.2 on the Jaccard index across 11,500 real queries (Grossman et al., SIGIR 2026).
- 85.7% of citations point to sites the brand does not own, and about 80% come from about 18% of domains (Żatuchin, June 2026).
- YouTube (22.9%) and Reddit (18.5%) lead the top-50 cited sources in AI Overviews (Ahrefs, September 2026).
- 97% of llms.txt files received zero requests in May 2026 logs from 137,210 domains (Ahrefs, June 2026).
- 51% of B2B software decision-makers start research with an AI chatbot more often than with Google, up from 29% a year earlier (G2, April 2026).
- The share of consumers who used an AI tool to find a local business in the previous 30 days rose from 9% in 2025 to 52% in 2026 (SOCi, September 2026).

### What Changed in 2026

Much of the advice circulating under the [generative engine optimization](https://metavert.io/generative-engine-optimization) label rested on a handful of early figures. The table lists the ones that have been superseded; none should be cited as current.

| Claim that circulated | What 2026 evidence shows | Source |
| --- | --- | --- |
| Superseded: adding quotations lifts visibility 41%, statistics 33%, fluency 29%, citing sources 28% (Princeton GEO paper, 2023) | Re-measured on ten modern engine families, the levers move citation on none. Earlier benchmarks had already found most such rewrites ineffective or harmful to retrieval. | Bajemon and Rochet, Sep 2026; C-SEO Bench, 2025; SAGEO Arena, KDD 2026 |
| Superseded: lower-ranked pages gain most, with rank-5 sites up 115% (same 2023 paper) | Not reproduced. C-SEO Bench found that ranking higher in the model's context works better than content rewrites. | C-SEO Bench, NeurIPS D&B 2025 |
| Superseded: 76% of AI Overview citations come from Google's top 10 | 37.9% rank in the top 10, 31.2% at positions 11 to 100, 31.0% beyond 100. BrightEdge reports about 17% with a different method. | Ahrefs, Mar 2026; BrightEdge, Feb 2026 |
| Superseded: AI cites content that is 25.7% newer | 75% of cited pages were updated within the last year; the freshness "is being manufactured by updates, not by new publishing". | Seer Interactive, Jul 2026 |
| Superseded: schema markup boosts AI citations | Difference-in-differences against 4,000 matched controls: AI Overviews −4.6%, AI Mode +2.4%, ChatGPT +2.2%, the latter two within noise. | Ahrefs, May 2026 |

A survey of 45 GEO studies reached the general conclusion: "No reviewed technique shows a stable, longitudinal, cross-platform causal effect on organic discoverability" (Martinez, July 2026).

### Retrieval and Query Fan-Out

Ahrefs attributes the fall in top-10 overlap to [query fan-out](https://metavert.io/query-fan-out): "the initial query is split into multiple related sub-queries. The pages that appear most often within those sub-query SERPs then get cited." A peer-reviewed study agrees on direction: 53% of the domains AI Overviews consult are not in the organic top 10, and two runs two months apart shared only 18% of pages against 45% for organic results (Kirsten et al., ACL Findings 2026).

How often an AI Overview appears depends on the query. Semrush measured 15.69% of 10 million keywords in November 2025, after a peak of 24.61% in July 2025. Xu, Iqbal and Montgomery found 13.7% across 55,393 queries but 64.7% for question-form queries (May 2026). Seer Interactive reports 95.4% of comparison queries, 85.9% of question-format, 36% of informational and 5% of transactional (2026, 5.47 million queries).

The traffic effect cuts both ways. An AI Overview is associated with a 58% lower click-through rate for the top organic result, up from 34.5% in April 2025 (Ahrefs, February 2026), while being cited in the overview is associated with 120% more organic clicks per impression than not being cited (Seer, 2026). Both are associations, not controlled effects. Google said at I/O in May 2026 that [AI Mode](https://metavert.io/google-ai-mode) "has surpassed one billion monthly users", and since 27 January 2026 the "Show more" control on an AI Overview opens AI Mode.

### Engines Disagree, and Answers Vary Daily

Each engine draws on a different slice of the web. Beyond the Grossman overlap figure, a small vendor-authored sample of 589 citations found 96.4% of cited URLs in only one of four engines (Tannenbaum, September 2026). Engines also use sources differently. Mean citations per prompt were 6.88 for ChatGPT, 12.06 for Google and 16.35 for Perplexity, while mean per-page influence on the answer ran the other way: 0.27, 0.058 and 0.065 (Zhang, He and Yao, April 2026). The authors separate "selection", getting cited, from "absorption", shaping the answer.

The same engine disagrees with itself over time. Day-to-day Jaccard similarity of cited sources averages 0.34 to 0.42, and 0.45 to 0.59 for brands mentioned, so visibility should be described "as a distribution rather than a single-point outcome" (Schulte et al.). Sielinski (March 2026) adds that citation distributions follow a power law and "many apparent differences between domains fall within the noise floor of the measurement process." Query language explains 26.5% of variance in one single-author study (Żatuchin, July 2026). Goodie counted 16 unannounced changes to social sourcing in seven months (September 2026).

Positions are also less settled than rankings imply. Only 15.2% of 1,094 categories have a clear ChatGPT "owner" and 53.7% are unsettled, although clear owners held first place 90.4% of the time month over month (Semrush and Growth Memo, July 2026). Tracking a brand's own position against these benchmarks requires repeated sampling across engines; tracking visibility per engine is the province of AI-search visibility tools like [LLM Optimizer](https://metavert.io/llmopt).

### Who Gets Cited

Most citations go to [earned media](https://metavert.io/earned-media), though the size of the owned share depends on the query set. Żatuchin measured 14.3% owned across 128 brands and 12 markets, with Wikipedia the top source in 11 of 12 languages. Kumar (June 2026; vendor author; 149,912 citations) put brand-owned domains at 2.9% and "best-of" listicles at 21.0%. The University of Toronto GEO study found earned media at 72 to 92% against 18 to 27% for brand-owned (September 2025). Local queries are the exception: Yext found first-party websites supply 44% of local citations and listings 42% (October 2025).

Brand mentions correlate with visibility more strongly than links do. Across 75,000 brands, YouTube mentions correlated at about 0.74 with ChatGPT and AI Mode visibility and branded web mentions at 0.656 to 0.709, against 0.266 to 0.326 for Domain Rating (Ahrefs, December 2025; Spearman correlation, not causation). Incumbency matters too: Kumar reports day-one visibility of 73% for household brands, 44% for mid-market and 11% for niche brands, and NanoKnow (SIGIR 2026) finds models "more accurate when answers were seen during pre-training", an advantage retrieval reduces but does not remove.

| Engine | Leading cited domains (share among top-50 cited sources) | Measured |
| --- | --- | --- |
| Google AI Overviews | YouTube 22.9%, Reddit 18.5%, Facebook 10.1%, Google 8.8%, Wikipedia 4.0% | Ahrefs, Sep 2026 |
| ChatGPT | Reddit 16.8%, Wikipedia 7.0%, YouTube 2.5% | Ahrefs, Sep 2026 |
| Perplexity | Reddit 21.6%, YouTube 20.8%, Wikipedia 6.3% | Ahrefs, Sep 2026 |
| Grok | Reddit 16.3%, YouTube 15.1%, Facebook 13.9% | Ahrefs, Jun 2026 |

These shares are not stable. Promptwatch data published by Semrush, labelled provisional, showed [Reddit](https://metavert.io/reddit)'s share of all ChatGPT citations falling from 3.8% to 0.5% in mid-August 2026; that figure uses a different denominator from the Ahrefs table and the two are not comparable. On recency, Seer found 75% of cited pages updated within a year (Gemini 78%, ChatGPT 73%, Perplexity 65%), and Zhen et al. fitted citation half-lives of about 39 days for high-timeliness queries and 68 days for low.

### Video and Social

Video citations are overwhelmingly long-form. Otterly's correlational study of more than 100 million citation instances found long-form at 94% of YouTube citations and Shorts at 5.7%; views and subscribers showed no relationship (r = −0.03 each), description length a modest one (r = 0.31), and 40.83% of cited videos had under 1,000 views (March 2026). Gumshoe, comparing 5,000 cited with 207,494 uncited videos, found that semantic similarity between the prompt and the title and description outperformed a model using all video and channel metadata (September 2026). BrightEdge reports that half of AI Overview queries citing a YouTube URL point to a specific timestamp (August 2026). Volatility applies here as well: YouTube's share of all Google AI Mode citations fell from 13.5% on 1 September 2026 to between 3.4% and 3.9% daily from 16 September (Otterly). See [video GEO](https://metavert.io/video-geo).

Social platforms' share of all AI citations rose from 4.9% to 7.2% between January and August 2026, with LinkedIn growing from roughly 8% of pooled social citations to about a fifth (Goodie, September 2026). Google AI Overviews cited Facebook in 19.5 million answers, Instagram in 877,000 and TikTok in 78,000 (BrightEdge, July 2026). Access explains part of the gap: robots.txt files checked on 28 September 2026 show tiktok.com and instagram.com disallowing the OpenAI, Anthropic and Perplexity crawlers while allowing Googlebot and Bingbot.

### Crawlers and Access

[AI crawlers](https://metavert.io/ai-crawlers) now divide by purpose. OpenAI documents GPTBot for training and OAI-SearchBot for its search index, both honouring robots.txt, and says of ChatGPT-User that "robots.txt rules may not apply" because the fetch is user-initiated. Anthropic runs ClaudeBot, Claude-SearchBot and Claude-User, all honouring robots.txt (Search Engine Roundtable, February 2026). Cloudflare Radar reported user-action crawling up "over 15x" in 2025, with training crawling at seven to eight times the volume of search crawling. From 15 September 2026 Cloudflare blocks training and agent bots by default on ad-supported pages for new domains, while search bots stay allowed. Publishing these controls is increasingly a platform feature: [LightCMS](https://metavert.io/lightcms), the CMS serving this site, exposes separate switches for training, AI search and user-fetch crawlers.

Google is the hard case. Google-Extended controls whether content is used for Gemini training and grounding, and under a UK Competition and Markets Authority order Google is testing a Search Console toggle to opt a site out of AI Overviews and AI Mode, UK-first (TechCrunch, June 2026). Blocking has a measurable cost: Grossman et al., comparing Google organic results, AI Overviews and Gemini, found sites blocking Google's AI crawler were less likely to be retrieved by AI Overviews.

For [llms.txt](https://metavert.io/llms-txt), the Ahrefs log study found AI retrieval bots made 1.1% of requests and coding agents and agent infrastructure 10.5%, the largest AI category. Mintlify, a documentation vendor, reports that agents made up 66% of traffic to its hosted developer docs in July 2026, up from 15.2% at the start of the year.

### Buyers and Coding Agents

In G2's survey of 1,076 B2B software decision-makers, 69% chose a different vendor than planned based on AI guidance and 33% bought from a vendor they had not known. Ten Speed, a vendor, found that evaluation-stage answers cite product pages (24.1%), comparison content (13.3%), listicles (13.2%) and G2 or Capterra (7.2%) (September 2026). Similarweb found people shown a brand in a ChatGPT recommendation were 2.5 times more likely to visit within seven days, with 55.9% of those visits arriving through branded search (via Search Engine Journal, June 2026). Tally reports 43% of new users arriving via AI assistants (September 2026) and Webflow about 10% of signups (November 2025); both are self-reported.

[AI coding agents](https://metavert.io/ai-coding-agents) are a separate channel. JetBrains found 90% of professional developers using them at work at least weekly (August 2026). Armature, a vendor, observed across 5,292 sessions that agents agreed on the same tool 42% of the time, and that Codex ran a web search in 94% of sessions against roughly 30% for Claude Code (September 2026). Amplifying found Claude Code wrote custom code instead of choosing a tool in 12 of 20 categories (February 2026). A University of Zurich study found six of ten provider-affiliated models significantly favour their own ecosystem, by up to 18.8 percentage points (May 2026).

### Local Search

Adoption of [local AI search](https://metavert.io/local-ai-search) is high and trust is conditional. BrightLocal found 40% of consumers trust AI platforms for business recommendations (February 2026). In SOCi's September 2026 survey, 67% had been given wrong information about a local business and 81% verify before contacting one. Inclusion is narrow: across more than 350,000 locations of multi-location brands, "ChatGPT recommends just 1.2% of brand locations, compared to an average 35.9% appearance rate in Google's local 3-Pack" (SOCi, January 2026), and Sterling Sky counted 5,943 unique businesses in AI local packs against 18,330 in regular 3-packs (June 2026). Whitespark's panel of 47 experts, an opinion survey, ranked presence on expert-curated "best of" lists as the top AI-visibility factor (November 2025).

### What the Evidence Does Not Yet Show

- **A reliable lever.** No published technique has a stable, cross-platform causal effect on citation. The brand-mention and video findings are correlations.
- **Comparable denominators.** Studies count share of top-50 domains, share of all citations or share of answers. The Reddit figures above differ thirtyfold for that reason alone.
- **That citations are accurate.** 11.0% of about 98,000 atomic claims in AI Overviews were unsupported by the cited pages (Xu et al.), and deep-research agents scored 39 to 77% on factual accuracy despite more than 94% valid links (Onweller et al., May 2026).
- **Resistance to manipulation.** FORGE showed one polluted page can alter up to 27% of recommendations; a proposed defence cut attack success from about 50% to 6% (GEO Defender, September 2026).
- **Conversion and growth rates.** No current primary source was found for widely repeated figures on AI-search conversion rates or crawler growth, so none appear here.
- **Independents.** The largest local datasets cover chains, not independent businesses.

### Method Note

This digest is a synthesis of published third-party studies, each opened and checked between 28 September and 6 October 2026. It contains no original measurement. Most industry figures are vendor-reported and are labelled as such where the author sells a related product; peer-reviewed venues are named. Figures carry dates because citation shares move within weeks. More on method is at [AI visibility measurement](https://metavert.io/ai-visibility-measurement), and the full page index is at the [GEO hub](https://metavert.io/geo).

## Related Topics

- [GEO: Generative Engine Optimization Guide](https://metavert.io/geo) — hub for every related page
- [Generative Engine Optimization](https://metavert.io/generative-engine-optimization) — the discipline this evidence bears on
- [AI Citations](https://metavert.io/ai-citations) — selection versus absorption explained
- [Query Fan-Out](https://metavert.io/query-fan-out) — why citations left the top 10
- [Earned Media](https://metavert.io/earned-media) — where most citations point
- [AI Visibility Measurement](https://metavert.io/ai-visibility-measurement) — sampling a result that changes daily
- [AI Crawlers](https://metavert.io/ai-crawlers) — training, search and user-fetch bots
- [Video GEO](https://metavert.io/video-geo) — what the YouTube citation studies support
- [Local AI Search](https://metavert.io/local-ai-search) — the local findings in depth

## Further Reading

- [AI Overview citations and the organic top 10](https://ahrefs.com/blog/ai-overview-citations-top-10) — Ahrefs, March 2026
- [AI Overviews one year on: presence, size and citing](https://www.brightedge.com/resources/weekly-ai-search-insights/ai-overviews-one-year-presence-size-citing) — BrightEdge, February 2026
- [Kirsten et al. on AI Overview source stability](https://aclanthology.org/2026.findings-acl.526.pdf) — ACL Findings, 2026
- [Xu, Iqbal and Montgomery: AI Overview prevalence and claim support](https://arxiv.org/abs/2605.14021) — arXiv, May 2026
- [Grossman et al.: Google organic results, AI Overviews and Gemini compared](https://arxiv.org/abs/2604.27790) — SIGIR 2026 / arXiv, April 2026
- [Tannenbaum: citation overlap across four engines](https://arxiv.org/html/2609.22655) — arXiv, September 2026
- [AI Overviews study](https://www.semrush.com/blog/semrush-ai-overviews-study/) — Semrush, November 2025
- [AIO impact on Google CTR: 2026 update](https://www.seerinteractive.com/insights/aio-impact-on-google-ctr-2026-update) — Seer Interactive, 2026
- [AI Overviews reduce clicks: update](https://ahrefs.com/blog/ai-overviews-reduce-clicks-update) — Ahrefs, February 2026
- [Search at I/O 2026](https://blog.google/products-and-platforms/products/search/search-io-2026/) — Google, May 2026
- [Scoring Without the Engine](https://arxiv.org/abs/2609.07559) — Bajemon and Rochet, arXiv, September 2026
- [Survey of 45 GEO studies](https://arxiv.org/abs/2607.14035) — Martinez, arXiv, July 2026
- [SAGEO Arena](https://arxiv.org/abs/2602.12187) — KDD 2026 / arXiv, February 2026
- [C-SEO Bench](https://arxiv.org/abs/2506.11097) — NeurIPS Datasets and Benchmarks 2025 / arXiv, June 2025
- [GEO: Generative Engine Optimization (original paper, for history)](https://arxiv.org/abs/2311.09735) — arXiv, November 2023
- [Does schema markup increase AI citations?](https://ahrefs.com/blog/schema-ai-citations/) — Ahrefs, May 2026
- [llms.txt request-log study](https://ahrefs.com/blog/llmstxt-study/) — Ahrefs, June 2026
- [Zhang, He and Yao: citation selection and absorption](https://arxiv.org/abs/2604.25707) — arXiv, April 2026
- [Don't Measure Once](https://arxiv.org/abs/2604.07585) — Schulte, Bleeker and Kaufmann, arXiv, April 2026
- [Sielinski: noise floors in citation measurement](https://arxiv.org/abs/2603.08924) — arXiv, March 2026
- [Żatuchin: query language and measurement variance](https://arxiv.org/abs/2607.13304) — arXiv, July 2026
- [Social media AI citations study 2026](https://higoodie.com/blog/social-media-ai-citations-study-2026/) — Goodie, September 2026
- [ChatGPT topic authority study](https://www.semrush.com/blog/chatgpt-topic-authority-study/) — Semrush and Growth Memo, July 2026
- [Żatuchin: owned versus third-party citations in 12 markets](https://arxiv.org/abs/2606.25787) — arXiv, June 2026
- [Kumar: 149,912 citations across five engines](https://arxiv.org/html/2606.20065) — arXiv, June 2026
- [Generative Engine Optimization: How to Dominate AI Search](https://arxiv.org/abs/2509.08919) — University of Toronto, arXiv, September 2025
- [AI citations study of 6.8 million citations](https://www.yext.com/about/news-media/ai-citations-release) — Yext, October 2025
- [AI brand visibility correlations across 75,000 brands](https://ahrefs.com/blog/ai-brand-visibility-correlations) — Ahrefs, December 2025
- [NanoKnow](https://arxiv.org/abs/2602.20122) — SIGIR 2026 / arXiv, February 2026
- [Most-cited domains in AI Overviews](https://ahrefs.com/blog/most-cited-domains-ai-overviews/) — Ahrefs, September 2026
- [Most-cited domains in ChatGPT](https://ahrefs.com/blog/most-cited-domains-in-chatgpt/) — Ahrefs, September 2026
- [Most-cited domains in Perplexity](https://ahrefs.com/blog/most-cited-domains-perplexity/) — Ahrefs, September 2026
- [Most-cited domains in Grok](https://ahrefs.com/blog/most-cited-domains-grok/) — Ahrefs, June 2026
- [Reddit's citations in ChatGPT fall](https://www.semrush.com/blog/reddits-citations-in-chatgpt-fall/) — Semrush (Promptwatch data), August 2026
- [Content recency's impact on AI visibility in 2026](https://www.seerinteractive.com/insights/study-content-recencys-impact-on-ai-visibility-in-2026) — Seer Interactive, July 2026
- [Zhen et al.: citation half-lives](https://arxiv.org/abs/2607.15771) — arXiv, July 2026
- [YouTube AI citation study 2026](https://otterly.ai/blog/youtube-ai-citation-study-2026/) — Otterly, March 2026
- [AI search updates: September](https://otterly.ai/blog/ai-search-updates-september/) — Otterly, September 2026
- [YouTube citation analysis](https://gumshoe.ai/resources/whitepapers/youtube-citation-analysis/) — Gumshoe, September 2026
- [How Google, OpenAI cite YouTube videos differently](https://www.mediapost.com/publications/article/417417/how-google-openai-cite-youtube-videos-differently.html) — MediaPost (BrightEdge data), August 2026
- [Google AI cites Facebook in 19.5 million AI answers](https://www.brightedge.com/news/press-releases/research-google-ai-cites-facebook-in-19-5-million-ai-answers) — BrightEdge, July 2026
- [Overview of OpenAI crawlers](https://developers.openai.com/api/docs/bots) — OpenAI, checked October 2026
- [Anthropic updates its crawler docs](https://www.seroundtable.com/anthropic-updates-its-crawler-docs-40978.html) — Search Engine Roundtable, February 2026
- [Radar 2025 Year in Review](https://blog.cloudflare.com/radar-2025-year-in-review/) — Cloudflare, December 2025
- [Content Independence Day: AI options](https://blog.cloudflare.com/content-independence-day-ai-options/) — Cloudflare, July 2026
- [Publishers will be able to opt out of AI search thanks to new regulation](https://techcrunch.com/2026/06/03/publishers-will-be-able-to-opt-out-of-ai-search-thanks-to-new-regulation/) — TechCrunch, June 2026
- [State of docs traffic](https://www.mintlify.com/blog/state-of-docs-traffic) — Mintlify, July 2026
- [Half of B2B software buyers now start their research with AI chatbots](https://www.prnewswire.com/news-releases/new-g2-research-half-of-b2b-software-buyers-now-start-their-research-with-ai-chatbots-302742807.html) — G2, April 2026
- [What AI cites at the B2B evaluation stage](https://www.tenspeed.io/blog/what-ai-cites-b2b-evaluation-stage) — Ten Speed, September 2026
- [AI-recommended brands saw 2.5x more site visits](https://www.searchenginejournal.com/ai-recommended-brands-saw-2-5x-more-site-visits-similarweb/580241/) — Search Engine Journal (Similarweb data), June 2026
- [6 years in, 6 million far](https://blog.tally.so/6-years-in-6-million-far/) — Tally, September 2026
- [Traffic is no longer reliable](https://www.growthunhinged.com/p/traffic-is-no-longer-reliable) — Growth Unhinged, November 2025
- [AI coding agent adoption 2026](https://blog.jetbrains.com/research/2026/08/ai-coding-agent-adoption-2026/) — JetBrains, August 2026
- [Which tools coding agents install](https://armature.tech/blog/which-tools-coding-agents-install) — Armature, September 2026
- [Claude Code picks](https://amplifying.ai/research/claude-code-picks) — Amplifying, February 2026
- [Ecosystem preference in provider-affiliated models](https://arxiv.org/pdf/2605.28515) — University of Zurich, arXiv, May 2026
- [Local Consumer Review Survey 2026](https://www.brightlocal.com/research/local-consumer-review-survey/) — BrightLocal, February 2026
- [Local Discovery Index: the AI local search verification loop](https://www.soci.ai/blog/ai-local-search-verification-loop/) — SOCi, September 2026
- [Local Visibility Index: in AI-driven discovery, few brands are chosen](https://www.soci.ai/news/in-ai-driven-discovery-few-brands-are-chosen-most-disappear/) — SOCi, January 2026
- [The state of local SEO in 2026](https://www.sterlingsky.ca/the-state-of-local-seo-in-2026/) — Sterling Sky, June 2026
- [Local Search Ranking Factors 2026](https://whitespark.ca/local-search-ranking-factors/) — Whitespark, November 2025
- [Onweller et al.: factual accuracy of deep-research agents](https://arxiv.org/abs/2605.06635) — arXiv, May 2026
- [FORGE](https://arxiv.org/abs/2606.13610) — arXiv, June 2026
- [GEO Defender](https://arxiv.org/abs/2609.02964) — arXiv, September 2026